Current topics in the assessment of retail mergers in the UK – the examples of Ladbrokes/Coral and Celesio/Sainburys

ACE Conference – 17 November 2016

Bojana Ignjatovic – RBB Chris Jenkins – CMA Ivan Olszak – CMA Diana Jackson – CRATomaso Duso – DIW

1

Overview

● Background

● National vs local effects and nature of competition

● Metrics of competition

● Online vs brick-and-mortar

2

Motivation

● Retail mergers form a central part of CMA (and other NCA) casework

● Ladbrokes/Coral and Celesio/Sainsbury’s were both phase 2 cases, running in parallel in first half of 2016

● Useful to compare/contrast the approaches taken

● Cases illustrate some of the key issues that regularly come up in retail merger cases

3

Ladbrokes/Coral – The parties overlapped in three product markets

Licensed Betting Offices (‘retail’)● Ladbrokes: 2,154 LBOs● Coral: 1,850 LBOs

4

Digital gambling services (‘online’)● Ladbrokes: -£24m

EBITDA in 2015● Coral: -£40m EBITDA in

2015

Operation of greyhound tracks● Ladbrokes: Crayford

and Monmore Green● Coral: Hove and

Romford

We will focus on the overlap in the retail market, but we will also discuss our assessment of the interaction between retail and online suppliers

Ladbrokes/Coral – The retail market is dominated by four national chains

UK LBOs gross gambling yield by segment (2014/15)

5

Share of LBOs by operator(March 2016)

Source: UK gambling commission, CMA calculations

Celesio/Sainsburys – Parties overlap in retail pharmacy

6

Lloyds (part of Celesio)● 1,540 retail pharmacies● Mainly located on High Streets or

in GP surgeries

Sainsbury’s● 277 pharmacies● Located in large format

supermarkets

As a result of the merger, Celesio will operate Sainsbury’s in-store pharmacies under the Lloyds brand. Sainsbury’s will continue to sell general sales list (non-prescription) medicines

Celesio/Sainsbury’s – more limited concerns at national level; focus on local

7

RetailerRetail pharmacy

market share (%)*Market share of

NHS revenue (%)†Independent/other 44 43Pharmacy chains: Boots [20–30] [20–30]Lloyds [10–20] [10–20]Well [5–10] [5–10]Rowlands [0–5] [0–5]Superdrug [0–5] [0–5]Total larger operators 44 49Supermarket pharmacies: Tesco [0–5] [0–5]Sainsbury’s [0–5] [0–5]Asda [0–5] [0–5]Morrisons [0–5] [0–5]Big 4 supermarkets 12 8Combined Lloyds/Sainsbury’s 14 16

Source: Verdict UK pharmacy report (2015).* Calculated on the basis of percentage of licences.† Calculated on the basis of sales revenue.

Overview

● Background

● National vs local effects and nature of competition

● Metrics of competition

● Online vs brick-and-mortar

8

Theories of harm should capture all possible effects on competition

● The traditional theories of harm in retail mergers are pretty straightforward:

- If the parties flex some parameters of competition locally (or if it would be profitable to do so), we ask whether the merger might affect incentives in local areas

- If the parties apply all parameters of competition uniformly across all shops, we ask whether the aggregation of local changes might affect incentives at the national level

● But are we not missing something with this traditional framework?

- Dynamic effects: what if the parties are expanding rapidly and tend to target the same types of areas?

- Different pricing mechanisms: what if prices are partly determined through auctions or bargaining processes (rather than Bertrand competition)?

- Innovation: what if innovation depends on the number of participants, or if one of the parties is particularly innovative?

9

In Ladbrokes/Coral, we considered both local and national theories of harm

How competition works

● Some parameters were flexed at the local level (local discounts, store refurbishment, staffing)

● Other parameters were set at the national level and applied uniformly across the estates (odds, return-to-player, etc)

● One aspect of competition (competition for the ‘top price’) involved an auction-like pricing mechanism distinct from standard Bertrand competition

10

Theories of harm

● Traditional theories of harm- Loss of competition at the local level

- Loss of competition at the national level (as a result of the aggregated loss of competition in local areas)

● Alternative theories of harm- Loss of potential competition, in areas

where the parties would have entered and competed against each other

- Loss of competition for the ‘top price’, for selections for which the parties were the most competitive bidders

- Loss of innovation, in case innovation depends on the number of suppliers

In Celesio/Sainsburys, key issue was the nature of local competition

11

● Regulation constrains parameters of competition- Prescription medicines either free, or sold at a regulated fixed price- Regulations set minimum standards of service- Licences constrain opening hours (minimum core hours typically 40

hours/week)

● But still evidence of competition on location and QRS- Evidence that these parameters drive customer choice- Evidence that QRS is generally set above the minimum levels, and lots of

variation in QRS at local level

● Are supermarket pharmacies different? - Different shopping missions, but parameters of customer choice are similar

• Some consumers only visit the in-store pharmacy and make no other purchases

- Survey diversion ratio estimates suggest that consumers see high street pharmacies and supermarket pharmacies as substitutes

Do Lloyds and Sainsbury’s compete locally?

12

● Agreement that pharmacies are able to make QRS decisions at local level, and strong incentive to maximise prescription volumes

● Issue of national policies vs local implementation – e.g. staffing levels, waiting time targets - National policy ≠ no competition at local level?

● How to deal with limited evidence of actual competition - Quantitative analysis inconclusive because of data limitations- Qualitative evidence of competition between Lloyds and supermarket

pharmacies

Diana JacksonNovember 2016

Evidence on flexing PQRS: betting shops

13

Diana JacksonNovember 2016

Evidence on flexing PQRS: pharmacies

Community pharmacies in general:• Significant entry impact

on volumes within 1.4/1.6 miles (urb/rur)

• Significant impact of multiple stores on waiting times in urban areas

• Significant impact of independent rivals on opening hours

• No relationship between margin and concentration

• Independent entry reduces average time to refurb from 7.1 to 3.4 years (within 0.2 miles, significant at 1%)

Supermarket pharmacies in general:• No impact of entry on

volumes• No impact on waiting

times• No impact on opening

hours• No relationship between

margin and concentration

• Supermarket entry reduces average time to refurb from 7.1 to 4.7 years (within 0.2 miles significant at 10%, no impact over wider catchments)

Sainsbury’s specifically:• No impact of entry on

volumes• No impact on waiting

times• No impact on opening

hours• No relationship between

margin and concentration

• No impact of Sainsbury’s entry on refurbishment (but small sample)

14

National vs. local effects and nature of competition

• National and local dimensions of competition play an important role– Hard to separate this step from the definition of the parameters of competition

– Mix of qualitative and quantitative analyses as a screening device 

• Prices are not the only (main) relevant dimension of competition – Prices are set nationally (Ladbrokes ) or are regulated (Celesio)

• Alternative parameters are fundamental to understand the nature of local competition – Discounts, quality and speed of service, opening hours, stocking levels and waiting times, 

share of the number of prescriptions, refurbishment,…

– They do seem to affect costumers’ choice – empirical evidence of different nature (survey, demand estimation, diversion ratios,…)

• Main Issues– More difficult to measure  Risk is to focus on the level and dimensions of competition for 

which we have better data

– More fundamentally, we do not have a clear theoretical understanding of their impact on (consumer) welfare

• Need to carefully understand, model, and estimate demand

15

National vs. local effects and nature of competition

• Some additional open challenges when looking at retail:

– Horizontal (competition for local costumers) vs. vertical effects (bargaining power vis a’ vis local/nationalwholesalers) Potentially relevant only for Celesio

– Dynamic effects (innovation, repositioning, entry/exit…) Partially discussed in both cases

But, too difficult to make accurate predictions?

– Competition from online shops See later

16

Overview

● Background

● National vs local effects and nature of competition

● Metrics of competition

● Online vs brick-and-mortar

17

The appropriate competition metric depends on the context

18

● How to construct the competition metric:- Should we use a fascia count or a store count?

- Should we weight the stores (according to distance? To other factors?)

- Do we need to use a different metric in areas where the parties have more than one store?

- Can we extrapolate survey results from a small number of areas?

● How to use it:- As a filter to identify areas that warrant a more in-depth assessment?

- As a mechanistic rule to identify areas where the transaction is problematic?

In Ladbrokes/Coral, all the evidence pointed to geography being key

19

● Key findings:- the competitive interaction between two

LBOs is heavily influenced by the distance between these two LBOs relative to the distance to other LBOs

- customers do not perceive strong differences between the LBOs of the national chains

● In addition:

- independents exert a weaker constraint

- the relationship between distance and competitive interaction is non-linear

- LBOs interact primarily with LBOs located within 400m

- the closest LBO is the strongest constraint

● Evidence considered:- Entry/exit analysis

- Customer survey (incllocal diversion ratios)

- Price-concentration analysis (re local discounts)

- Analysis of local shop refurbishment decisions

- Internal documents and third party views

So we designed a new competition metric, the WSS, that we used mechanically

20

● Steps to estimating the Weighted Share of Shop (WSS) for a given LBO:

- Assign a weight to all LBOs around the centroid LBO based on distance

- Adjust the weight for the closest shop (x1.2)

- Adjust the weight for independents (x0.9)

- Divide the sum of the weight(s) assigned to the other party’s LBO(s) by the sum of all weights

● We used this metric to identify areas that failed the test ‘mechanically’

● We only did a couple of adjustments for areas with very low or very high densities of shops

Weights applied to LBOs

Worked example of WSS calculation

In Celesio/Sainsburys, we developed a similar competition metric

21

● Used share of stores rather than fascia count - Brand not a strong driver of pharmacy choice; large share of independents

● Shares weighted by distance- Store gets weighting of zero at edge of catchment, one at the centre, and linear relationship in between- Reflects strong empirical relationship between distance and expected rate of diversion (Fig 1)- Tested different weighting approaches - linear weighting gave best fit with survey diversion ratios (Fig 2)

● Asymmetric catchment issue – how to deal with wider supermarket catchments? - For a given distance, supermarket pharmacy has a higher weight than high street pharmacy

Fig 1: Relationship between survey diversion estimate and distance between surveyed stores

Fig 2: Relationship between weighted share of stores and survey diversion estimates

But took a more conservative approach with the filter and then considered individual areas

22

● Conservative first stage approach - filtered in overlap areas where Parties had- Combined share of stores greater than 40%, with increment of 15%- Parties closest or ‘closest but one’, with combined share of stores greater than 30% and

increment of 10%- Diversion predicted by demand estimation model greater than 25%

● (Overarching question of complexity vs making use of all the available evidence)● Qualitative second stage approach

- Used survey case studies to determine characteristics of local areas where Parties were particularly close competitors

- Formulated qualitative rules- Applied these to maps of local areas filtered at the first stage

● Why the difference in approach compared with Ladbrokes/Coral? - Differing information/data available in the two cases – approach has to fit the evidence

available- Difference in number of overlap areas and practicalities of carrying out detailed local

assessment in large number of areas

Diana JacksonNovember 2016

Evidence for measures of local competitionBetting shops:• Fascia versus stores: Fascia not

“sufficiently reflective of the evidence”• Distance: Evidence on PQRS flexing• Proximity: 1.2 weight on closest rival

“Primarily based on the CMA survey” – then regressed diversion ratios on different versions of the WSS to find the best fit

• Strength: 0.9 weight on independents based on:

• Feedback from rivals and evidence that independents are struggling

• Econometric evidence on entry/exit • Survey evidence on LBO impact

Pharmacies:• Fascia versus stores: fascia counts

perform best at predicting survey diversion for England…

• Distance: Linear shape gives best fit of WSS to survey diversions

• Proximity: Strong relationship between survey diversion and distance

• Strength:• Different catchments informed by

Lloyds and Sainsbury’s average 80% catchments by urbanicity

• No impact of prescription density• Boots strength and Superdrug

weakness not quantified: part of qualitative assessment

23

Diana JacksonNovember 2016

Testing measures of local competitionBetting shops:

• Positive association but low R2

(slightly higher including all LBOs)• WSS understates concerns at low

DR/overstates at high?

Pharmacies:

• Positive association and better R2

(higher again for England only)• WSS overstates concerns at low DR?

24

Diana JacksonNovember 2016

Setting thresholds for local competition

• Thresholds differ depending on fascia count• Overlaps within 400m: 35% combined

WSS and 5 to 4 store count or worse• Overlaps within 1600m: 35%

combined WSS and 2 to 1 store count

• No minimum increment to WSS

• Thresholds definitive

Betting shops:

• Thresholds differ depending on whether closest and measure of diversion used• WSS > 40% and increment > 15%• WSS > 30% and increment > 10%

where closest or second closest• Demand estimation diversion > 25%• Survey diversions > 30% (or 25%?)

• First three thresholds set to be conservative (regulation, asymmetry):

• Survey diversion applied at final step (therefore not conservative?)

Pharmacies:

25

Diana JacksonNovember 2016

Evidence for setting thresholds

• “Critical” survey diversion threshold based on gross margin and GUPPI (15-20% diversion, 10-20% GUPPI)

• “Candidate” WSS thresholds set based on relationship between WSS and survey diversion• 35% WSS “equivalent to” 17.7%

diversion

• Cross-check: do LBOs failing threshold fall into categories where there is evidence of a relationship between concentration and PQRS?

Betting shops:

• Financial analysis of gross margin variations with volume losses due to opening hours reduction (but less emphasis in final decision):• 40% (provisional findings)• 65% (+ sample bias correction)• 30% (+ fully variable staff costs)• Wide range (internal diversion)

• Refer to 15% DR used in supermarkets• 40% WSS said to be consistent with

30% DRs (chart suggests ~25%?)• No comparison with local data on actual

QRS effects

Pharmacies:

26

Metrics of competition

• At the local level: geographic market definition play a crucial role– Demand‐side substitutability is key: distance/location to the shops, preferences for brands, 

specialization,…

– Best case scenario: understand, specify, estimate demand  diversion ratios

– If not, collect other information

1. Define catchment areas 1‐mile radius around each of the relevant pharmacies

10/15‐minute drive‐time in urban/rural areas

Area including 80% of the pharmacy’s prescription customers

400m and 800m radii

Entry‐exit analysis: distances up to 1,600m

27

1.4m

4.7m

Metrics of competition

2. Measure the “intensity of competition”: Fascia counts vs. weights vs. distance vs. strength

– WSS seems reasonable pragmatic approach but quite ad hoc, which leavesquestions oni. the robustness of the findings provide results for extreme cases

ii. comparability across different cases

• Again: Need to carefully understand, model, and estimate demand

28

Overview

● Background

● National vs local effects and nature of competition

● Metrics of competition

● Online vs brick-and-mortar

29

Relationship between online and bricks and mortar a growing issue in many retail mergers

30

● Eg. see Wiggle/CRC phase 1 case – first UK case between two online retailers

● Online not a major issue in Celesio/Sainsbury’s- Online pharmacies in the UK have very small market share, mainly because of

regulatory constraints

- Interesting question about whether this channel will grow in future – but not a key issue for the merger investigation

● Much more significant issue in Ladbrokes/Coral- Large and growing online gambling sector

- Key question = how far online gambling constrains bricks and mortar LBOs

In Ladbrokes/Coral we used a range of evidence to assess the online constraint

31

● Price differentials between the two channels The parties argued that the dynamics of competition were changing rapidly, in that more and

more retail customers were regarding online suppliers as good substitutes

If this was true, then we should observe a narrowing of the price differential between the two channels over time

There was some evidence that the differential had been compressed slightly, but this change was modest in magnitude and did not apply to all product lines

● Survey evidence Face-to-face surveys indicated much lower levels of online diversion than telephone or

online surveys

The parties argued that face-to-face surveys underestimated the ‘true’ online diversion as they did not weigh customer responses by spend and they involve a ‘framing bias’

● Migration vs diversion question The parties argued that gradual customer migration from brick-and-mortar shops to online

suppliers proved that online constrains brick-and-mortar

This argument confuses migration and diversion, which have very different implications for suppliers incentives

Online vs. Brick and mortar competition

• It will become crucial for many retail sectors

– Considered only in Ladbrokes

– But relevant also for pharmacies: e.g. EU CJ decision for Germany

• It affects both market definition and competitive assessment

– We still do not know how to integrate competition from online

• Again! We need to better understand demand/substitution

– In Ladbrokes a good starting point

– Very little empirical evidence on this

32

Source: Hortaçsu and Syverson, Journal of EconomicPerspective, 2015

Online vs. Brick and mortar competition

• See the billions prices project @MIT “[…] my findings imply little within‐retailer price dispersion, both online and offline. While the Internet may not have reduced dispersion across retailers, it seems to have created the incentives for companies to price identically across their own physical and online stores. More research is needed to understand the mechanisms that drive this effect”

Cavallo: “Are Online and Offline Prices Similar? Evidence from Large Multi‐Channel Retailers,” AER, forthcoming

33

http://bpp.mit.edu/

Online vs. Brick and mortar competition

• Use rich household data and estimate demand – mostly in the marketing literature“[…] households are more brand loyal, more size loyal and less price sensitive in the online channel. […] Our research confirms the complementary nature of the online store to offline stores. For many households, the online store is an extension of the physical stores that has more flexible shopping hours and alleviates the burden of grocery shopping.”

Chu et al.: “An Empirical Analysis of Shopping Behavior Across Online and Offline Channels for Grocery Products: The Moderating Effects of Household and Product Characteristics,” Journal of Interactive Marketing, 24 (2010) 251–268

34